Skip to content

Latest commit

 

History

History
61 lines (50 loc) · 2.19 KB

README.md

File metadata and controls

61 lines (50 loc) · 2.19 KB
Raw

Sample functions

Sampler which counts the number of times we find ourselves in a particular function. Outputs a list of functions with their counts.

Usage:
  usample <start_bbcount> <end_bbcount> <bbcount_interval> [<filename>]
  Parameters:
    start_bbcount: Time, given as a basic block count to start sampling.
    end_bbcount: Last time that may be sampled.
    bbcount_interval: How often to take a sample, in basic block count.
    filename: A file to output the sampled stacks to.

  E.g. 1 1000 1
  means sample every basic block count from 1 to 1000.

Before using the script it must be loaded in to the debugger:

source PATHTOADDONS/sample_functions/sample_functions.py

Generating flame graphs

The output from this tool can be used directly to generate flame graphs. These graphs provide a visual indication of the busiest areas of your program. Generally, around 1,000 samples will produce a useful flame graph. You may wish to limit the sampling to a specific range, for example after start up has completed.

To generate a flame graph:

  • clone the flame graph repository

  • run the usample script. In this case we first query the recorded time range within UDB:

> info time
Current time is: 1 (in recorded range; [1 - 140,988,372])

To get around 1,000 samples we wish to sample every 141,000 bbcounts, so we run usample with:

> usample 1 140988372 141000 /tmp/stacks.data
  • in the shell run flamegraph.pl on the generated call stack data:
$ flamegraph.pl /tmp/stacks.data > /tmp/stacks.svg

You can then view the generated SVG file in your browser. Clicking on a block will allow you to zoom in to see more detail. You can also search for function names with regular expressions.

Run flamegraph.ph --help to see more options for controlling the output.

These flame graphs show the proportion of basic blocks that run in a particular call stack. This is related to execution time, but is not identical. Not all basic blocks take the same amount of time to run. In addition, this doesn't show time spent in different processes, in particular time spent in system calls.